This is a submission for the GitHub Finish-Up-A-Thon Challenge
https://github.com/YohannesAH/Hand_Written_Recognize
What I Built
This is a handwritten digit recognition app where users draw Arabic numerals on an HTML canvas and click predict to identify them. The drawing is converted into a base64 image and sent via a POST request to a Flask backend. The backend preprocesses the image (grayscale conversion, thresholding, contour detection, noise filtering, dilation, and resizing) to match MNIST dataset format before passing it to a pre-trained ML model, which then predicts the digit(s).
Demo
https://github.com/YohannesAH/Hand_Written_Recognize/blob/main/docs/assets/demogif/sample.gif
The Comeback Story
My idea for the unfinished part of the project is to integrate Azure AI text-to-speech with multilingual support for the recognized digits. I also plan to improve scalability by using a serverless architecture and deploying the application on AWS using Kubernetes.
My Experience with GitHub Copilot
GitHub Copilot supported my process by helping me write and refine code faster. It suggested useful boilerplate, function structures, and helped debug small issues in both JavaScript and Python. It also helped me understand different implementation approaches, especially when building the Flask backend and preprocessing pipeline for the ML model. However, I reviewed and adjusted all suggestions to make sure they fit my project requirements and logic
Note:CNN(Convulational-Neural-Network-CNN-) https://github.com/vzhou842 was the sole contributor "https://github.com/vzhou842" which I cloned, modified, and retrained for this implementation.
Top comments (0)